Oct 23rd 2025|5 min read
Listen to this story
Elly Ntonde was revising for his chemistry exams in Budondo, Uganda. The village has unpaved roads, no running water and flickering electricity. Yet when the 18-year-old was struggling with how metals react to acid, the world’s most advanced tutor was only a few taps away. He walked to a shop, bought 100mb of data and loaded it onto his phone. In seconds, ChatGPT had explained the answers.
Chart: The Economist
Mr Ntonde’s experience captures the promise of artificial intelligence. Less than three years after ChatGPT’s launch, some 800m people—a seventh of the world’s adults—use it each week. Many are in the developing world, where populations are young and tech-savvy. After America, India and Brazil are the biggest markets. A UN survey finds confidence in AI is higher in countries where human-development scores are lower. And, according to GWI, a consumer-research firm, Ghanaians and Nigerians are among the keenest users.
Can AI democratise knowledge—putting a tutor, doctor or adviser in every pocket? Early studies hint at the promise. In Nairobi, OpenAI and Penda Health, a chain of primary-care clinics, tested a tool that advised doctors during consultations. In a randomised trial covering nearly 40,000 patient visits across 15 clinics, doctors with the assistant cut diagnostic errors by 16% and treatment errors by 13%. In Nigeria, a six-week after-school scheme using Microsoft Copilot—in which pupils interacted with the chatbot twice a week—boosted English scores by the equivalent of nearly two years’ extra schooling.
The hope is that, like mobile phones before it, AI might bypass old bottlenecks. In the 1990s most African countries had fewer than one telephone line per 100 people. By skipping the wire stage and going straight to mobile, they reached near-universal phone access within two decades. AI could spread through cheap smartphones and local models. To do so, three hurdles must be cleared: connectivity, users’ abilities and institutional capacity.
Begin with connectivity. AI requires internet access. Although nine in ten people in rich countries were online in 2024, only one in four were in poor countries. Nearly 85% of Africans live within reach of a mobile-broadband signal. But data, even in pay-as-you-go form, is often too pricey.
The good news is that, from the user’s point of view, AI is relatively cheap. A single search-results page, laden with images and ads, uses 3,000 times more data than a text-based AI query. Owing to falling inference costs, sending a prompt to ChatGPT was already 90% cheaper in 2024 than loading search results. That could make access to information more affordable. Yet users still have to be online; efforts to deliver AI services via SMS remain prohibitively expensive, as mobile operators impose hefty mark-ups. Until data becomes cheaper and connectivity expands, the AI revolution risks missing the poorest.
Even where people are connected, many lack the skills to use AI productively—the ability hurdle. The World Bank estimates that 70% of ten-year-olds in low- and middle-income countries cannot read a simple text. For new users, opening a chatbot, typing a prompt or interpreting its response can be daunting. Getting value from such tools also requires knowing what to ask. Nicholas Otis of the University of California, Berkeley, and co-authors found that stronger Kenyan entrepreneurs raised their profits by more than 15% with an AI assistant, while weaker ones saw profits fall after following generic advice. In Budondo, Mr Ntonde sees a similar divide. Roughly half of young people own basic smartphones and experiment with AI, he says, but most use it primarily for entertainment—such as creating Ghibli-style portraits of themselves to post on social media—rather than to study or work.
Language magnifies the problem. Most AI systems are trained mainly on English and other rich-country tongues, leaving hundreds of African languages barely represented. The result is a gulf between what AI can say and what many can understand. A small but growing group of researchers is trying to close the gap. Projects such as Masakhane, Ghana NLP and Kencorpus—community-led initiatives that build open datasets for African languages—are laying the groundwork. Open-source and voice-based tools now complement these efforts, hinting at a future where people can speak to machines in their own languages.
But the highest hurdle is not access to technology. Iqbal Dhaliwal of the Abdul Latif Jameel Poverty Action Lab, a research centre, notes a history of “silver-bullet” technologies that failed since they were not integrated into existing institutions. Massive open online courses, hailed as the future of education, barely lifted learning outcomes in poor countries as they operated outside schools—offering content without the teachers or exams to make it stick. AI could now follow a similar path. Taha Barwahwala of Columbia University and co-authors studied a model used in one Indian state to detect fake firms. Although the algorithm successfully flagged thousands of non-existent companies, enforcement failed to improve as officials lacked incentives to act on its findings.
Ultimately, the success of AI will depend on whether it can lift productivity across entire economies, not just improve individual services. As Lant Pritchett of the London School of Economics notes, no country has achieved mass education or good health without first becoming richer. Broad-based growth, driven by rising worker productivity, is what underpins lasting gains in human capital.
And technologies raise productivity only if businesses reorganise to exploit them. When factories first replaced gas lamps with light bulbs, little changed; when they redesigned production around electric machinery, output soared. Research by Diego Comin of Dartmouth College and Martí Mestieri, then of Northwestern University, looked at 25 general-purpose technologies over two centuries and found that newer inventions such as PCs and the internet reached poor countries faster, yet their use remained shallow. AI adoption may be still more demanding. Even in rich countries, firms struggle: in America only about one in ten report using the tech in production. For poorer economies, the challenge is steeper still. ■
For more expert analysis of the biggest stories in economics, finance and markets, sign up to Money Talks, our weekly subscriber-only newsletter.
Artificial intelligenceFinance & economics
This article appeared in the Finance & economics section of the print edition under the headline “Leaps and limits”
Discover stories from this section and more in the list of contents
Subscriber only | Money Talks
Delivered to your inbox every week
Yes, I agree to receive exclusive content, offers and updates to products and services from The Economist Group. I can change these preferences at any time.